Module 1 Assignment: Experiment with Text Continuation Styles

Objective: Create a Colab notebook that uses GPT-2 to generate creative text continuations with different styles.

Requirements:

• Load a pre-trained GPT-2 model (using HuggingFace transformers - same approach as used in GPT-2.ipynb)
• Create 3 different story starters in different genres/styles. For example:
  • Fantasy/Adventure: “In a land of dragons and magic…”
  • Sci-fi: “The year is 2157. Humanity has just…”
  • Mystery: “The detective examined the crime scene and noticed…”
  • (or choose your own three)
• Then adjust for:
  • Greedy decoding vs. sampling
  • Different temperature values
  • How the opening sentence shapes the continuation (e.g., short vs. long)
• Document your observations (using Markdown in the notebook)
  • What differences do you notice between the strategies?
  • What worked well (or surprised you!)
  • What didn’t work that well

Deliverable: A Colab/Jupyter notebook with:

• Code cells with your implementation
• Outputs of generated text from GPT-2
• Markdown cells explaining what each sampling strategy does and any observations

Hint:

For better results, use a larger GPT-2 model on Colab T4.

tokenizer = GPT2Tokenizer.from_pretrained("gpt2-medium")
model = GPT2LMHeadModel.from_pretrained("gpt2-medium")